Ballet Form Training Based on MediaPipe Body Posture Monitoring

People are increasingly turning to the cloud in the context of “healthy China” to engage in online exercise. The use of artificial intelligence technology to address broad population health-related challenges has become increasingly important as information technology has matured. The MediaPipe artificial intelligence framework, which Google recently released, is used in this article to optimize video feedback and support the “cloud movement” of widespread home ballet instruction in order to examine the effects of digital technology-enabled ballet training on the general improvement of physical health. Based on the experiment’s findings, trainers can use MediaPipe video feedback as an additional method of “cloud movement” training in public homes. This allows trainers to visually reflect on any issues that arise during the exercise process and to promptly modify training according to motion monitoring in order to reduce the risk of physical injury brought on by improper motion training. It is possible to envision a day in the future when video feedback built on the MediaPipe architecture would combine artificial intelligence with exercise training to achieve training objectives that enhance the accuracy of trainers’ motions and enhance physical balance.


Introduction
Today, various cloud fitness-assisted apps and physical products are born as a combination of artificial intelligence and sports fitness, providing reference value for the use of digital technology and the reconciliation of fitness [1][2][3][4], such as Keep, fitness "magic mirror," black technology, and so on.Ballet training is a new form of sports training that combines artistic and normative behavior, and it is the favored fitness program for young people, according to recent research and analysis.However, improper fitness movements can result in physical impairment, so research and methods to encourage the public to perform sports exercises with the correct movements at all times without professional guidance and "cloud fitness" at home to enhance health are crucial.
With the advancement of technology, the calculation of human gesture assessment in the field of AI technology has increased the monitoring accuracy of human body joint nodes [5,6].Gait analysis is the scientific study of the gait cycle, which plays a significant role in the diagnosis of musculoskeletal and neurological disorders and in evaluating the efficacy of various interventions given to patients.It uses Mediapipe human posture monitoring [7].Kim et al. [8] utilized Mediapipe technology to detect and warn of the elderly's collapse.Based on AI technology, Baranyi et al. [9] developed a system of knee rehabilitation exercises to assist patients in completing physical therapist-determined exercise programs.Latreche et al. created a website for remote Rehabilitation and Motion Studies on Mediapipe for patients with right hip fractures.Raju et al. [10] used the new three-dimensional human gesture estimation technology based on prospective target detection (HPEM) in motion video analysis with success.Motion counting has been augmented with artificial intelligence technology by Sikdar et al. [11].Palani et al. [12] provided real-time joint angle estimates based on the Mediapipe framework and inertia sensors for the analysis of human motion techniques.Kwon and Dongho [13] utilized posture detection during exercise in real time.
Based on the MediaPipe framework, the human gesture assessment application is intended to provide ballet form trainer learning, combined with video feedback, to apply it to the public's fitness and physical health improvement, regulate fitness training movements, enhance the effect of sports, and promote the development of the general population's fitness and health system.This article will select subjects with the aid of MediaPipe for a ballet form training experiment, select the selected Plié standing and Pas sauté down motion in the two ballet shapes, and analyze the body-specific joint data, such as the angle of bending the knee joint, two legs, and other parameters during the movement.The result will be timely feedback to the subject, and the experimental results demonstrated that the use of MediaPipe framework assisted ballet training has improved the performance of the subjects.This article, from a multidisciplinary perspective, integrates research, the application of computer vision and artificial intelligence technology to the fields of art, sports training, and other fields, using the computer to extract relevant data, the data for mathematical statistics and analysis, to discover the problems of movement in training, optimize video feedback in people's applications of ballet "cloud fitness", and update the means and methods of people at home to exercise.

Human posture monitoring methods
The most intuitive way to detect human gestures is by distinguishing gesture movements by man, such as the number of complex movements used in competition videos, which is accurate but takes a lot of time and human effort.With AI technology spreading into people's lives, the current stage of video monitoring technology can easily monitor the target character and track the movement trajectory of the character, further development can monitor the specific behavior movements of the person [14].

MediaPipe Human Behavior Monitoring
This article utilizes Google's MediaPipe, a cross-platform, open-source framework, to estimate the human joint coordinates in each frame image.MediaPipe employs machine learning (ML) to construct pipelines and process video data, inferring 33 3D landmarks and background masks from the RGB video framework on the entire body.The MediaPipe gesture monitoring module divides the human body into 33 points, as shown in Figure 1.As long as the data of the specified points in the model is output, the correct human body part can be determined for each point.From RGB video screens, MediaPipe inferred 33 body-wide 3D landmarks and backdrop separation masks.In Figure 1, it can be seen that the MediaPipe gesture monitoring module divides the human body into 33 points, and in the model, the output of the data of the specified points can be derived for the correct body parts.

Calculation of characteristics
MediaPipe uses 33 main nodes to monitor posture.To determine the pose feature parameters, the coordinate points are multiplied by the pixel values.The distribution of human bones and limb movements during relevant activities are used to develop computation models for 33 important nodes of the human body.In squatting, the horizontal angles of the shoulder, knee, ankle, and legs are taken into account.For adjacent skeletal angles, the angles between the two lines are calculated by geometry analysis.The function calculates the angle of each human joint.For example, to calculate the right knee bending angle, the three-dimensional coordinates of joint points A, B and C are extracted as shown in Figure 3.We can assume that AB and BC are bone structures, with a straight line AB with BC intersecting at point B, and the angle of clutch AB with BC can be determined by the following Formula (1), construction of vector , BA BC    Included angle cosine: (2) When calculating the physical balance parameters, the overall volatility is reflected according to the average value, the standard deviation and the range of the measurement part data.
The average value: 1 Range:

Plié standby movement norms and physical balance
The Pulley Squat was explained and demonstrated in order to guide the experimenters to achieve maximum outward opening of the legs to "shoulder width" without compromising the vertical position of the body or the correct stance of the feet.When squatting, the legs should be open, the core should be taut, the waist should not collapse, the back should be perpendicular to the ground, the palms of the feet should be flat against the floor, and the movement should be at an even pace.After describing and instructing the movement, the experimenters' mobile phones were set up to capture the video.In order to increase the generalizability of the data and reduce the possibility of error, experimenters were instructed to record three groups of Plié squats each time, one of five movements, and to provide feedback on the recorded video.The data were statistically analyzed following the experiment, as shown in Figure 4, which depicts the histograms of movement normative parameters and body equilibrium parameters pre-, during-, and post-experiments.Before they saw the video processed by the MediaPipe framework video feedback, experimenters with no dance background demonstrated variations in the angle of the legs during squatting, nonstandard squatting movements, and unbalanced postures for the pre-experimental parameters.

Post
The experimenter was guided to focus on the shoulder joints, leg angles, knee joints and angle, and ankle joints during squatting, as well as to practice and correct any errors, using the MediaPipe frame video feedback.

, , s k a
A A A represent the average angles of the shoulder, knee and ankle.
, , represent the standard differences in the angle of the shoulder, knee and ankle.
, , represent the range values of the angles of the shoulder, knee and ankle.
After three months of training with MediaPipe video feedback, the experimenter's squatting stability and the standardization of their squatting movements were significantly enhanced.In order to compare the data before and during the experiment, the results are graphed in Figure 4, which demonstrates that the performance and the stability of the squat improved, and that the values stabilized.

Pas sauté small movement specification and body balance
The experimenter was guided by an explanation and demonstration of the pas sauté leap, and based on the experimenter's specific conditions and practice of the plié jump, the experimenter is trained in the second pas sauté jump movement.Pas sauté small jump is a standing jump.During the standing, the knees are relaxed, the flexibility of the knees is used and care is taken to control the posture in the air to ensure height and falseness during the jump.The legs are straight in the air without gaps, the knees are tightened and the legs are externally rotated.When landing, the toes, the sole, and then the instep smoothly stick to the ground, focusing on a light and graceful landing.Focus points for the Pas sauté movement specification include the three elements of the leap preparation gesture, the air gesture, and the landing gesture, and the angle of the legs must be monitored.This was also the case with Plié, where subjects recorded three sets of pas sauté jumps, consisting of eight movements on each take, and then gave feedback on the recorded video.As depicted in Figure 5, the parameters and physical equilibrium parameters for the experimenter's pre-, during-, and post-experimental movements are specified as follows.After three months of ballet training based on the MediaPipe framework video feedback assistance Subjects, the subjects showed a certain improvement in the angle of air gestures and legs during the Pas sauté small jump movement.This indicates that subjects have had the awareness of the movement normative with the help of MediaPipe frame video feedback, and consciously tightened their legs during the jumps.There was also a good improvement in the strength of both legs that the subjects had built up during the exercise.The movements that the subjects completed during the experiment were improved, stability was enhanced and the values levelled off.Experiments have shown that MediaPipe framework video feedback can help experimenters improve the completion of movements, and enhance the experimenter's ability to control the body and movements.

Conclusion
Ballet form training, as one of the family "cloud fitness" options, requires training with standardised and scientific movements, in order to achieve the training objective of improving physical fitness and avoiding sports injuries caused by incorrect movement training.To address this problem, MediaPipe video feedback can be used as an auxiliary tool for the "cloud exercise" at home, visually reflecting the movement problems presented by trainers during the exercise, and helping to detect and make training adjustments in a timely manner.By repeatedly watching the MediaPipe videos to compare the movements with the standard movements or to analyse the problems with the movements, users can avoid the difficulty in spotting details of body movements and proactively correcting them in time by only watching instructional videos and videos recorded on their mobile phones during practice.Ballet training using the MediaPipe framework can be clearly visualized, the combination of Video Feedback with Ballet Training has a strong possibility and feasibility, and the data feedback is also intuitive.

Figure 3 .
Figure 3. Implementation of Schedule Action normative parameter chart (b) Physical balance parameters Figure 4. Pre-, during-and post-experiment Plié standby movement norms and physical balance parameters diagram Action normative parameter chart (b) Physical balance parameters Figure 5. Pre-, during-and post-experimental Pas sauté small jump movement specifications and physical balance parameters diagram

Table 1 .
Table 1 lists body part node parameters.Related body parameters.

) 4 . Mediapipe human gesture monitoring under ballet form training experimental process
This experiment selected Plié standing and Pas sauté jumping in the two ballet shapes with the clearest body angle and joint point presentation state in order to experimentally examine the movement specification and physical balance.Three months of MediaPipe-based video feedback training to evaluate pre-and post-experimental movement.